1. What is the fundamental difference between previous augmentation strategies and the learned augmentation strategy proposed in AutoAugment and RandAugment? (1-2 lines)
 
2. What is the main advantage of using RandAugment method over previous methods of learned augmentation like AA, Fast AA, etc. ? (1-2 lines)

3. Why selecting a small-proxy task might lead to learning sub-optimal data augmentation policies? (Explain in 2-3 lines)